5 minute read
Startup Story

Cyber superheroes: Overwatch and AI turn the tables on cyberattackers

The Overwatch Data team pose together indoors for the camera.

The open source data intelligence platform uses ML to stay ahead of bad actors and disruptive events

Zara Perumal had been a software engineer on Google’s Threat Analysis Group for four years when fellow Googler Arjun Bisen approached her with a compelling proposition: to do what she was already doing, but make it publicly available. It was the perfect opportunity for Zara to apply her skill set and Google’s machine learning technology “to build software systems at scale that would have a geopolitical and real-world impact.”

Zara, Arjun, and Tad Mielnicki went on to cofound Overwatch Data, an open source data intelligence platform using machine learning to understand and surface relevant information and events across news, social media, deepweb, and natural disasters that pose risks or opportunities to enterprise companies. Zara brought her engineering expertise to her role as CTO, applying artificial intelligence to unearth cyber threats on a global scale.

Executive protection for notable figures, financial fraud targeting companies, exploitation of natural disasters, increasing ethnic tensions, and cyber attacks impacting the critical infrastructure of public schools and hospitals: “all of that is in our interpretation of online or digital risk,” Zara explains. “Our mission is to take this open information and make it relevant for people who are trying to protect their users or their company.” Overwatch demystifies the global threat landscape with transparent data that gives its users a clearer view of trends and dangers that could impact them. Zara continues, “We're protecting infrastructure from cyber attacks to build a safer world for everyone—and specifically, we help decision makers better protect their infrastructure by having the information they need to make informed decisions.”

Artificial intelligence is at the center of Overwatch’s technology and is necessary for performing at scale. “There's so much data,” Zara tells us, “more than any person can process. So we use AI to cut through the noise and find those needles in the haystack that are risks for our customers' businesses—and then we use more AI to make it actionable, relevant, and personalized.” AI is used at every step in Overwatch’s process and, owing to rapid advancements in AI, they can approach and solve problems in ways that weren't possible even as recently as a year ago.

“With these latest set of large language models, we can understand and get to the nuanced set of what's happening in the event,” she says. This means distinguishing between a threat that constitutes large-scale, real-world harm or a complaint against a single company or actor. “That fine-grained nuance is what those large language models are letting us do, and the whole infrastructure behind them, the technology for things like vector databases, is building a system that lets us process at scale in a way that you had to have years of work and a ton of computers to do as a startup before.”

While Zara “absolutely thinks AI is the platform of the future”, she also believes that there is an inevitable downside that makes it easier for “bad actors” to pursue harmful agendas. This is why, she says, companies like Overwatch have to exist. “I think that the only way to keep up with that base is to use AI on the defensive side. AI can streamline so many things, whether it's analysis, synthesis, understanding what's relevant, allowing users to access all this data that they can make sense of and use to inform their reports.” AI-based technology can also systematically improve itself as it learns and evolves, which Zara hopes will allow their models to improve detections as attackers themselves continue to evolve.

“We live in a world of understanding data in bulk,” Zara tells us. Building technology as advanced and security-critical as Overwatch required using tools that could handle many layers of complexity and carry heavy data loads. Overwatch began by hosting their web service using App Engine and Cloud Sql, and used Vertex AI tools for prototyping: “Both the AutoML tools and new LLM / Palm interfaces have let us prototype finding signals in data before moving to more custom models”, says Zara. Bigquery helped them to understand and process large amounts of data and they deployed Overwatch’s systems securely using Google’s security toolkit, which includes CloudIDS and SecretManager.

As a former Googler, Zara brought to Overwatch a comprehensive skillset in systems building and machine learning, but she was was excited to receive non-dilutive funding and additional product mentorship via the Google for Startups Latino Founders Fund. Zara and her cofounders benefited from their mentors’ breadth of expertise that was outside of their own domains, from IP and patents—“that was fascinating because none of us are lawyers”—to team building—“[to learn] from people who have gone through building a team in a remote-first environment and building that cohesion, that sense of mission without burning our team out, is super useful.”

As well as the technical expertise, the Latino Founder Fund provided Zara with a meaningful sense of community. Zara recognizes the culture of supportiveness and willingness to share that infuses the community, from her time at Google. It’s a huge benefit given that, she says, “a lot of the problems are more shared than they are unique. We are trying to figure out building a team and a culture that we like.” Many of the companies on the program are also using machine learning and building software systems, which has resulted in an abundance of sharing resources and experiences among the founders. “As a founder, it can sometimes feel lonely or like you're doing some crazy vision on your own that's a terrible idea, but having the community of people who are in it with you and are willing to have the candid conversations about how this is really hard, but we're all motivated about our work, is something that I really appreciate.”

The Latino Founders Fund also offered representation, which was missing from Zara’s experience as a founder. She tells us about her grandmother, who emigrated from Honduras and is “the hardest working person I know. And when I say standing on the shoulders of giants, she's five foot tall, but she's truly set the bar for her love of education and set me and my family up for success. On a personal level, that drives me every day. I try to live up to that.” Professionally, Zara tells us that she was previously used to feeling unseen: “I'm a minority and people usually don't assume I'm the CTO, whether that's because I'm Hispanic, or a woman or queer—or for other reasons.” As well as access to capital and expertise, the Latino Founders Fund aims to provide the belonging and visibility of being part of an ecosystem of founders with a shared sense of identity.

As for Overwatch’s goal of being, in her words, “the Netflix of enterprise risk,” Zara is confident about the company’s future as a leader in cybersecurity. “In this environment, the tech has exploded so quickly that if you're 10 years old, you don't have that much headway on us. I'm not as worried about that and more about whether can we move fast enough to solve this problem quickly enough while making something that's a really good experience.”

Learn more about Overwatch Data